Introduction
Lithium-air (Li-air) batteries represent a potential "holy grail" for electric vehicle energy storage due to a theoretical energy density comparable to gasoline. However, practical implementation is hindered by electrolyte instability. During discharge, these batteries produce superoxide radicals ($O_2^{\bullet-}$), aggressive nucleophiles that decompose organic electrolytes. This study, conducted by researchers at MIT and PNNL, addresses the challenge of identifying kinetically stable electrolytes through a combination of high-level quantum chemistry and automated computational screening.
Methodology
The research team utilized a two-pronged approach to evaluate electrolyte stability:
- User-Driven Benchmarking: High-accuracy manual calculations were performed on cyclic carbonates and sulfones using Density Functional Theory (DFT) for geometry optimization and DLPNO-CCSD(T) for final energy barriers.
- Automated Pipeline: An end-to-end computational pipeline was developed to scale the search for transition states (TS). This pipeline integrates SMILES/SMARTS reaction encoding, relaxed potential energy scans (PES), and nudged elastic band (NEB) calculations.
- Machine Learning Integration: A polarizable atom interaction neural network (PaiNN) was trained on over 1,000 generated geometries to evaluate the potential for ML-driven reaction path identification.
Kinetic Stability: Carbonates vs. Sulfones
The study reveals a stark contrast in the kinetic stability of different electrolyte classes against superoxide attack. Carbonates were found to be highly susceptible to nucleophilic attack, with reaction barriers typically $\le 26$ kcal/mol, translating to half-lives of only hours or days at room temperature.
In contrast, sulfones exhibited significantly higher activation barriers, generally ranging from 32 to 38 kcal/mol. These higher barriers suggest a kinetic half-life exceeding 5,500 years, marking sulfones as promising candidates for stable Li-air battery electrolytes.
Interestingly, the research debunked the assumption that increasing substitution always improves stability. In tetrasubstituted carbonates, steric strain from "eclipsed" substituents actually lowered the reaction barrier, making the molecules more reactive than their less-substituted counterparts.
Automation and Machine Learning Results
Automating the search for transition states remains a significant hurdle. The pipeline faced challenges with small, flexible radicals like superoxide, which often rotated into unintended "superoxide-catalyzed" orientations rather than the target $S_N2$ pathways. Despite these difficulties, the PaiNN potential showed promise, achieving a Mean Absolute Error (MAE) of 1.88 kcal/mol/Å for energy gradients.
Conclusion
This work establishes a framework for the high-throughput screening of electrolyte stability. While sulfones demonstrate superior kinetic resistance to superoxide attack, the path toward fully automated discovery requires further refinement of transition state search algorithms and the development of more robust interatomic neural network potentials. Future efforts will focus on reactive molecular dynamics to simulate real-time degradation across vast chemical spaces.
Summary updated